Study On Event-triggered Filtering For Delayed Neural Networks

With the continuously development of technology in computer,telecommunication and control systems,networked control systems have been widely applied in the production and life.Due to the limited network bandwidth,there are always some non-ideal problems,such as network-induced delay,packet loss,chaos to order when data are transmitted.The traditional control systems transmit data by periodic-triggered way and this transmission mode wastes a lot of bandwidth resources of the network.In order to solve this problem,scholars have put forward the event-triggered scheme,it makes the two trigger time interval before and after not less than one sampling period,and reducing the amount of data which need transmitted in the network.On the other hand,Neural network is widely used in many fields,because of the limitation of transmission medium of physical device,delay is one of the main causes of the instability of the neural networks.Therefore,the research on the stability and state estimation problem of delayed neural networks still has important theoretical and application value.Estimation/Filtering problem of delayed neural networks has a large number of results,but there is few research on networked delayed neural networks under the event-triggered scheme,many problems and results still need further study.This thesis studies the state estimation and filtering problem of three kinds of neural networks,the main research results include the following three parts:?1?For a class of recurrent continuous delayed neural network,the problem of state estimation under event-triggered scheme is studied.Considering signal quantization,the network-induced delay and the neurons delay,the system model is established as a time-delay system,based on the Lyapunov stability theory,free-weighting matrix method and delay segmentation method,put forward the co-design method of event trigger conditions and state estimator,a state estimator with asymptotic performance is obtained.?2?For a class of discrete delayed neural networks,the problem of L_?filtering based on event-triggered scheme is studied.Considering the delays of neurons and network transmission,the system model is established as a multiple time-delay system,considering the influence of missing output measurements at the same time,the Lyapunov-Krasovskii functional is established,using delay segmentation method,linear matrix inequality and Jenson inequality,the stability problem of filtering system is studied,and put forward a method to design a kind of L_?filter with an adaptive trigger parameter.?3?For the discrete neural networks with mixed delays,the L₂-L_?filtering problem based on event-triggered scheme is studied.Considering the sensor saturation,signal quantization,the system model contains the neuron delay,distributed delay and communication delay,by using Lyapunov-Krasovskii functional method,nonlinear processing technology,delay segmentation method,LMI processing technology,a sufficient condition of stability of the filtering error system with L₂-L_?performance is obtained,and then the L₂-L_?filter is designed.